[molpro-user] parallel running molpro 2009.1

[Manhui Wang]

Hi John,
     I have looked at your CONFIG file just now. It seems some arguments
are unnecessary. Do you really use slater
type orbital? If not, -slater is unnecessary. In addition, molpro
configure can pick up the right blas library arguments automatically. It
seems -var
LIBS="-L/usr/lib64 -libverbs -lm" is something for high speed
network library (like Infiniband). If you plan to run molpro on just one
box and
without such high speed network, this option is also unnecessary.

The configuration may be something like:
./configure -batch -icc -ifort -mpp -mppbase
/n/home07/jtravers/apps/MYMPI/mpich2-1.3a1/include -blaspath
"/n/sw/intel/mkl/10.1.1.019/lib/em64t
INSTBIN=/n/home07/jtravers/apps/MOLPRO/molpro2009.1.mpich2
-var INSTLIB=/n/home07/jtravers/apps/MOLPRO/molpro2009.1.mpich2/lib

This is for MPI-2 version of Molpro. In addition, you can also try the
Global Arrays version of Molpro, in which GA should be built firstly.

Please refer to the benchmark results for both versions at
http://www.molpro.net/info/bench.php?portal=visitor&choice=Benchmarks

Best wishes,
Manhui

Manhui Wang wrote:
> Hi John,
>     The parallel performance depends
> on many factors, such as communication speed, I/O, memory, and code
> itself. When running jobs with heavy I/O and/or large memory, the
> bottleneck may be the file system and/or system memory. In addition,
> some code in Molpro is partly parallelized. Please refer to the comment
> http://www.molpro.net/pipermail/molpro-user/2006-July/001867.html
> 
> Best wishes,
> Manhui
> 
> John Travers wrote:
>> Dear All,
>>
>>         We had some strange situations when running molrpo (2009.1 patch
>> 26) on a linux box with 2 quadcore Xeon E5410. molpro was compiled with
>> intel fortran 11.1, mpich2-1.3a1 (we also tried other versions but got
>> the same results) and MKL-10.1.1.019 (the configuration file is
>> attached). The compilation was successful. Running a simple test
>> calculation (close-shell CCSD(T) with a large basis set), we found the
>> parallel scaling stopped (especially the time for the election integral)
>> when requesting with more than 4 cpus. The mpich2 was compiled with
>> icc/ifort, and the communication device is ch3:nemesis. Any suggestion?
>>
>> Thank you!
>>
>> Best wishes
>>
>> J.T.
>>
>>
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> Molpro-user mailing list
>> Molpro-user at molpro.net
>> http://www.molpro.net/mailman/listinfo/molpro-user
> 

-- 
-----------
Manhui  Wang
School of Chemistry, Cardiff University,
Main Building, Park Place,
Cardiff CF10 3AT, UK
Telephone: +44 (0)29208 76637